Original Article
Comparative analysis of iterative reconstruction algorithms with resolution recovery for cardiac SPECT studies. A multi-center phantom study

https://doi.org/10.1007/s12350-013-9821-0Get rights and content

Abstract

Background

This investigation used image data generated by a physical phantom over a wide range of count statistics to evaluate the effectiveness of several of the newer commercially available SPECT reconstruction iterative algorithms (IRR) in improving perfusion defect contrast and spatial resolution, while controlling image noise.

Methods

A cardiac phantom was imaged using four different gamma cameras over a wide range of counts statistics (from 6 to 0.8 Mcounts). Images were reconstructed with FBP, OSEM, and the IRR available on site. IRR were applied without corrections (IRR NC), with attenuation correction (IRR AC), scatter correction (IRR SC), and attenuation + scatter corrections (IRR SCAC). Four image performance indices related to spatial resolution, contrast, and image noise were analyzed.

Results

IRR NC always determined significant improvements in all indices in comparison to FBP or OSEM. Improvements were emphasized with IRR SC and IRR SCAC. Count reduction from 6 to 1.5 Mcounts did not impair the performances of any of the considered indices.

Conclusions

This is the first study comparing the relative performance of different, commercially available, IRR software, over a wide range of count statistics; the additional effect of scatter and attenuation corrections, alone or in combination, was also evaluated. Our results confirm that IRR algorithms produce substantial benefits with respect to conventional FBP or OSEM reconstruction methods, as assessed through different figures of merit, in particular when SC and/or SCAC are also included.

Introduction

Recent software improvements in iterative image reconstruction with resolution recovery (IRR) allowed to halve the activity of injected tracer for MPI examinations using a standard acquisition time, leading to a decrease in patient’s as well as operator’s exposure1 in comparison to optimized protocols for filtered back-projection (FBP) reconstruction.2 These software developments take into account the main factors involved in the loss of quality of the myocardial perfusion images, namely the noise and the depth-dependent loss of resolution inherent in parallel-hole collimators.

The image noise depends, first, on the amount of events (counting statistics) registered in the projection matrices of the MPI study, and second, on reconstruction process starting from these matrices. The new algorithms are able to reduce the noise in the reconstructed images during the iterative process, with noise-suppression methods different among the software vendors. The IRR algorithms also use the physical characteristics of detector and collimators of the SPECT scanner and the patient-specific data of orbit shape, radius, or distance from patient to detector to improve image quality. Moreover, they offer the possibility of applying additional scatter (SC) and/or attenuation (AC) correction algorithms based on CT data for the integrated SPECT/CT systems.

All data corrections for image degrading factors have shown to improve image quality in cardiac SPECT reconstruction.3 However, their effectiveness could vary with the reconstruction algorithm used.4

Most scanner manufacturers have implemented a version of IRR algorithms into conventional software for ordered-subset expectation maximization (OSEM) iterative reconstruction. These IRR algorithms, developed for conventional SPECT cameras, include Astonish (Philips Healthcare, Eindhoven, The Netherlands),5 Evolution for Cardiac, EfC (GE Medical Systems, Waukesha, WI, USA),6 Flash3D (Siemens Medical Solutions),7 and wide-beam reconstruction, WBR, by a third-party vendor (UltraSPECT, Haifa, Israel).8

Several studies have been published since IRR algorithms were implemented in the clinical practice.9, 10, 11 The major aim of these studies was to evaluate the physical characteristics, the performances, and/or their effectiveness in clinical use, either halving the acquisition time or the tracer activity.

Since these algorithms are specifically adapted for each camera and collimator, few studies compared their relative performances. Published studies comparing different advanced IRR from different vendors, showed different performances depending on the figure of merit considered.12,13 However, these studies are generally restricted to a single point of the count statistics. Moreover, the comparison between different reconstruction methods or acquisition modalities was performed taking into consideration one figure of merit at a time.

The aim of this study was to characterize the relative performances of the abovementioned IRR algorithms over a wide range of acquisition count statistics, also in respect to conventional reconstruction software (OSEM and FBP), using an anthropomorphic phantom and selected physical figures of merit for image evaluation. The application of additional data emission corrections such as for scatter and attenuation and the performance of different combinations of scanner/software were also studied. A multivariable approach was used to take into account at the same time the relative effect of the different main factors (reconstruction protocols, scanner/software combinations, and count statistics), on the specific figure of merit, considered as dependent variables.

Section snippets

Multi-center Study Design

Four nuclear medicine departments (referred later as “centers”) participated to the multi-center study allowing the use of the following scanners: a BrightView (Philips), two Infinia (General Electric), and a Symbia-T2 (Siemens) cameras, each one equipped with their respective low-energy high-resolution (LEHR) collimators. The gamma-cameras characteristics and the manufacturers’ recommended protocols are detailed in Table 1. Three of four cameras were integrated SPECT/CT systems, while the

Phantom and Image Quality Evaluation

An example of the reconstructed image sets (best short-axis slices for PD phantom) for FBP, OSEM, and IRR (with and without AC/SC) from the 6 × 106 and 0.8 × 106 count scans of center N.4, is reported in Figure 3.

Figure 4 shows the three-dimensional surface graphs with distance-weighted least square fitting methods of LV wall thickness varying the count statistics and the reconstruction method for the four centers enrolled in the study. Figure 5 shows the values of PD contrast in LV wall

Discussion

Although the IRR algorithms are well introduced in clinical practice, a complete characterization of their relative performances at varying levels of count statistics of the acquisition scan, from clinical reference of standard MPI procedure to very low total counts, has never been carried out. To the best of our knowledge, this is the first study comparing the relative performance of different, commercially available, IRR softwares, over a wide range of count statistics; the additional effect

Limitations

Results from the present experimental study are obtained with an anthropomorphic torso phantom: this could not be necessarily applicable to clinical examinations. Further clinical studies on patients are needed to confirm the full portability of our results to the clinical ground, in particular to verify the lowest level of count density able to maintain the diagnostic and prognostic accuracy, expected to be in the range of one quarter of the standard SPECT acquisition.

Moreover, resolution

New Knowledge Gained

This is the first study comparing the relative performance of different, commercially available, IRR software, over a wide range of count statistics; the additional effect of scatter and attenuation corrections, alone or in combination, was also evaluated. Our results confirm that IRR algorithms produce substantial benefits with respect to conventional FBP or OSEM reconstruction methods, as assessed through different figures of merit, in particular SC and/or ACSC are also included.

Conclusions

The image reconstruction with the IRR algorithms produce better physical figures of merit than conventional FBP or OSEM algorithms. Their performances critically depend on the combined application of additional scatter and/or attenuation corrections that take full advantage of the capability of such algorithms. To a lesser extent, they also depend on the scanner/software combination. The impact of count statistics on the performances of IRR algorithms in cardiac SPECT images can be neglected

Disclosures

None.

References (17)

There are more references available in the full text version of this article.

Cited by (35)

View all citing articles on Scopus
View full text